Claude Cowork can scrape websites using plain English. No Python, no scripts, no setup beyond installing the desktop app and a Chrome extension. You describe what data you want, point it at a URL, and it extracts the information and saves it to a spreadsheet or slide deck on your computer.
This guide covers exactly how to set it up, which sites you are actually allowed to scrape, and two real examples: pulling sold eBay listings into a spreadsheet and scraping a sports card marketplace to build a slide deck. It also covers when Cowork scraping makes sense and when a different tool is the smarter call.
What Is Claude Cowork Web Scraping?
Claude Cowork is the desktop-only agentic mode of Claude that works directly with files and apps on your computer. On its own, it can read and write files, run multi-step workflows, and connect to tools like Google Drive and Notion. But it cannot browse the web on its own.
The web scraping capability comes from pairing Cowork with Claude in Chrome, an official browser extension that gives Claude direct control over a Chrome tab. When you enable this connector in Cowork, Claude can navigate to a URL, read the page, extract structured data, and save it to your computer all from a single plain English prompt.
The combination is genuinely no-code. Claude handles the browser navigation, DOM parsing, and data formatting. You just describe the output you want.
See also: Claude Cowork tutorial
What You Need Before Your First Scrape
Three things are required before Claude Cowork can scrape anything.
1. Claude Desktop with a Paid Plan
Claude Cowork only runs inside the Claude desktop app, and it requires a paid plan. Pro ($20/month), Max, Team, or Enterprise all qualify. The free tier does not have access to Cowork.
Download the app at claude.ai/downloads and log in with your Anthropic account. On Windows you will see a setup banner asking you to enable a Windows feature for the secure sandbox. Click Enable, confirm the PowerShell prompt, and restart. After the restart the banner disappears and Cowork is ready.
2. Claude in Chrome Extension
Go to the Chrome Web Store, search for Claude, and install the official extension. Pin it to your Chrome toolbar via the Extensions menu so it stays accessible. When you first install it, the extension will prompt you to log in and authorize the connection to your Claude account.
3. Enable the Claude in Chrome Connector in Cowork
Open the Claude desktop app, click Cowork, then click the plus icon and select Connectors. Find Claude in Chrome in the list and enable it. The connector enables browser tools inside Cowork sessions. Cowork will not take action on any site until you explicitly approve it, either for a single session or always.
Once those three things are in place you are ready to scrape. One important note: when Claude is controlling a Chrome tab, you will see a gold outline around that tab. That is the visual signal that Claude is actively working in the browser.
One practical tip from Anthropic’s own safety guidelines: consider using a separate Chrome profile for scraping work. This keeps Claude’s browser access isolated from your personal accounts, bookmarks, and saved passwords. If you are logged into banking, email, or work tools in your main Chrome profile, a second profile gives you a clean environment where Claude only sees what you explicitly open for it.
See also: Claude Cowork Projects
See also: Claude Cowork Dispatch guide
See also: Claude Cowork Excel guide
See also: Claude Cowork PowerPoint guide
Need Help Building AI Automations?
Join Our AI Community
What Sites Are Legal to Scrape and What to Avoid
This is the part most tutorials skip. Scraping the wrong website can get your account banned, block your IP, or create legal problems depending on what the site’s terms of service say. Before pointing Claude at any website, check two things: the robots.txt file and the terms of service.
How to Read a robots.txt File
Every website should have a robots.txt file at domain.com/robots.txt. This file publishes rules for what automated bots are allowed and not allowed to do on the site. It is used for SEO but it is also the first place to check before scraping.
The key fields to look for:
- User-agent: * means the rule applies to all bots, including Claude in Chrome.
- Disallow: /path means bots are not allowed to access that path. Do not scrape pages under a disallowed path.
- Allow: /path explicitly permits access even if a broader Disallow rule applies.
- A blank Disallow field or no Disallow entries means the site is generally open to crawling.
Some sites allow 60 to 70 percent of their pages while blocking the rest. Always scan the full robots.txt before deciding a site is safe to scrape.
Terms of Service: The Second Check
If you are logged into a website, the terms of service matter even more than robots.txt. A robots.txt file governs bots. The ToS governs account holders. Scraping while logged in and violating the ToS can get your account permanently banned.
LinkedIn is the most notorious example. Scraping LinkedIn while logged in is a direct ToS violation and accounts get removed for it. Do not scrape LinkedIn.
Sites Claude Cowork Will Not Scrape
Claude Cowork has built-in safety filters that block certain categories of sites regardless of what you ask. Financial services sites like Stripe’s homepage will be declined automatically. Cowork will tell you directly in the chat that the site is blocked.
Beyond the built-in blocks, avoid these categories:
- Social media platforms: Instagram, Facebook, Twitter/X. ToS violations and high ban risk.
- Financial data sites: brokerage platforms, payment processors, banking portals, investment platforms, and cryptocurrency exchanges.
- Paywalled content: any site where the content requires a paid subscription to access.
- Adult content and pirated content sites. These are blocked by Claude’s built-in content filters entirely.
- Any site where robots.txt has a Disallow rule covering the pages you want to scrape.
The General Rule of Thumb
If a website is publicly accessible without logging in, has no terms of service clause against scraping, and the relevant pages are not disallowed in robots.txt, it is generally safe to scrape. When in doubt, check the ToS and look for a scraping or data use clause. This is not legal advice, just practical guidance on where Claude Cowork will and will not work.
Prompt Injection: A Higher Risk During Scraping
Prompt injection is when malicious instructions are hidden inside a web page’s content, invisible to you but readable by Claude. When Claude is actively browsing a page to extract data, it is reading everything on that page. A site could theoretically include hidden text designed to redirect Claude into doing something unintended.
Anthropic has built safeguards specifically for this in Claude in Chrome, including content classifiers and action confirmations for anything high-risk. But the practical advice is simple: watch what Claude is doing during a scrape, only point it at sites you trust, and never leave financial accounts, email, or sensitive browser tabs open while Claude in Chrome is active. Use a separate Chrome profile for scraping work if you want a clean separation between Claude’s browser access and your personal accounts.
How to Scrape eBay Sold Listings with Claude Cowork
eBay’s sold listings are publicly accessible and not restricted in robots.txt for general browsing. This makes it a good target for market research, pricing analysis, or tracking what a specific category of items is actually selling for.
Here is the step-by-step process from start to spreadsheet.
Step 1: Create a Folder for Your Scraped Data
Before running any prompt, create a new folder on your computer for this project. Something like ‘ebay-scrape’ or ‘cowork-scrape-output’. In Claude Cowork, click the folder icon, choose ‘Select a different folder’, and pick that folder. This tells Cowork where to save the output files.
Step 2: Write the Scraping Prompt
Open a new Cowork session and write a prompt like this:
Replace [your search term] with whatever you are researching. The same structure works for vintage vinyl records, collectibles, electronics, or any eBay category with publicly listed sold items.
Being specific about the time range (last 30 days) and the sale type (auction vs Buy It Now) gives you more useful structured data rather than a raw list of listings.
Step 3: Authorize Chrome and Handle the Connection Error
When Claude initiates the browser task, you may see an error like ‘Chrome is not connected’ even if you have already set up the extension. This happens when the connection token has expired. Click Authorize in the error message, confirm the authorization in the Chrome extension popup, and then click ‘Try again’ in the Cowork chat. Claude will pick up exactly where it left off.
Step 4: Review the Results
For a search returning around 47 results, Claude took approximately 6 minutes to scrape the full page, check the DOM structure, and confirm each listing’s details. The output is a spreadsheet with these columns:
- Sold date
- Listing title
- Sale price
- Sale type: Auction (with number of bids) or Buy It Now
Claude also color-codes the entries automatically: yellow for auctions, green for Buy It Now, red for unknown sale type (usually Best Offer acceptances). A summary tab includes totals, averages, and median price across all confirmed sales.
See also: web scraping with Python
Scraping Across Multiple Pages
If the data you want spans more than one page of results, include pagination instructions in your prompt. Claude does not automatically continue to page 2 unless you tell it to.
Two approaches work well. The first is to tell Claude how many pages to scrape upfront:
The second approach is to let Claude decide when it has enough data by specifying a row target: ‘Keep scraping pages until you have at least 100 listings.’ For sites where the number of relevant results is unpredictable, this gives Claude flexibility without leaving you with incomplete data.
Keep in mind that each additional page significantly increases runtime and token consumption. For deep multi-page scrapes, a Python scraper is the more practical long-term solution.
Example 2: Scraping a Sports Card Site and Building a Slide Deck
The second example goes further: scrape multiple search terms from a different site and compile the top 10 results into a PowerPoint slide deck with card images included.
The target site was 130point.com, a sports card marketplace with publicly available sold data. The prompt searched for two different card sets, asked for the top 10 sales across both, and requested that each card’s image appear on its slide.
The Prompt
The key additions here compared to the eBay prompt are: multiple search terms in a single task, a slide deck as the output format instead of a spreadsheet, an image requirement, and a filter to exclude lot sales (multiple cards bundled together).
What Came Back
Claude needed clarification mid-task on whether to include lot sales. After confirming single cards only, it completed the scrape and generated a slide deck with the top 10 results. The runtime was over 30 minutes for two search terms plus the slide deck build.
The data was accurate. The slide layout was functional. The main issue was image quality: the card images were distorted and stretched on the slides. A better approach is to scrape the data first, confirm the results look right, and then ask Claude to build the slide deck in a separate step using the confirmed data.
Token Usage Reality Check
Three page scrapes in one session (one eBay search and two 130point searches) plus the slide deck build used 14 percent of the weekly usage limit on a $100 Claude plan. That is roughly 1 out of every 7 weekly scraping sessions before hitting the limit.
This matters if you are thinking about running this workflow daily or for many different search terms. Token consumption is the key trade-off with Cowork scraping.
Claude Cowork Scraping vs. Other Approaches
Cowork is not the right tool for every scraping job. Here is a direct comparison of when to use it versus the alternatives.
Use Claude Cowork When:
- You need to scrape a site once or occasionally. No setup time, no code to write.
- You want structured output (spreadsheet, slide deck, PDF) in the same workflow as the scrape.
- The site is simple with clearly structured listings and no aggressive anti-scraping measures.
- You do not know how to code and need results now.
Use a Python Scraper When:
- You need to scrape the same site regularly. A Python script runs without consuming tokens every time.
- You want full control over the scraping logic, rate limiting, and retry behavior.
- The site changes occasionally and you want to be able to update the scraper yourself.
If you want to build a Python scraper but are not sure where to start, Claude Code can generate the entire script for you. You write the prompt once, get working code, and that code runs on its own going forward without touching your API usage limits.
Use n8n or a Third-Party Scraper When:
- You need scheduled, automated scraping on a recurring basis.
- The site has aggressive anti-scraping measures that Cowork cannot handle.
- You are working with sites known to block standard scrapers (LinkedIn, large e-commerce platforms).
Tools like Apify and Firecrawl specialize in scraping at scale and already maintain scrapers for many high-demand sites. For sites where even a custom Python scraper would struggle, a service like Apify that maintains the scraper infrastructure for you is the more reliable long-term solution.
Frequently Asked Questions
Can Claude Cowork scrape any website?
No. Claude Cowork will not scrape sites that violate its built-in safety rules, including financial services platforms, payment processors, and sites with explicit anti-scraping ToS clauses. Sites protected by login walls and paywalls are also off-limits. For publicly accessible sites without scraping restrictions in their robots.txt file, Cowork generally works well.
How do I check if a website allows web scraping?
Go to domain.com/robots.txt and look for User-agent: * followed by Disallow entries. If the pages you want to scrape are not listed under Disallow, they are generally crawlable. Also check the site’s Terms of Service for any clause about automated data collection or scraping, especially if you have an account on the site.
How many tokens does Claude Cowork use for web scraping?
Scraping is token-intensive. In a live test, three page scrapes plus a slide deck build used 14 percent of the weekly limit on a $100 Claude Max plan. One-off scrapes are fine, but if you plan to scrape the same site regularly, a Python script or n8n workflow will be more cost-effective since they do not consume API tokens on each run.
What is Claude in Chrome and do I need it for Cowork scraping?
Claude in Chrome is an official Chrome extension that gives Cowork direct control over a browser tab. It is required for web scraping. Without it enabled as a connector in Cowork, Claude has no way to navigate to a website or interact with its contents. You install it from the Chrome Web Store and enable it in Cowork under Settings > Connectors.
What is the best alternative to Claude Cowork for recurring web scraping?
For recurring tasks, a Python scraper is the most cost-effective option since it runs without consuming API tokens. For scheduled automation, n8n workflows paired with a scraping library work well. For tough sites with aggressive anti-scraping measures, Apify and Firecrawl maintain managed scraping infrastructure that handles the technical complexity for you.
Next Steps
Claude Cowork web scraping is best treated as a one-off research tool, not a recurring data pipeline. For a single market research pull, a quick competitive pricing check, or extracting a page of listings you only need once, it is fast and requires zero setup beyond installing the Chrome extension.
The moment you need the same data weekly or the token cost starts adding up, the smarter move is a Python scraper or an n8n workflow. Both can be built with Claude’s help in a single session and they run independently from that point forward.
Start with one of the prompts from this guide. Scrape a site you have been meaning to research, check the robots.txt first, and see how far plain English takes you.
See also: Claude Cowork tips and tricks
